Percorrer por data de Publicação, começado por "2026-07-14"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Belief revision applied to profile dynamicsPublication . Almeida, Yuri Aristides da Silva Godinho de; Fermé, Eduardo Leopoldo; Nascimento, Arthur Henrique Casals do; Bermúdez I Badia, SergiBelief revision is the process by which an intelligent agent modifies its beliefs when confronted with new information that contradicts its current knowledge state. Whilst the AGM framework provides rigorous normative principles for rational belief change, a persistent gap exists between these theoretical foundations and practical implementation. Concurrently, user profiling has be come ubiquitous across recommender systems, adaptive interfaces, and cognitive rehabilitation platforms; however, the temporal dynamics of user profiles remain largely unaddressed by for mal, principled methods. Most systems resort to ad hoc recreation of profiles from scratch, and data-driven approaches suffer from recency bias, leading to volatile and opaque profile evolution. This thesis addresses three research questions. First, it investigates what insights can be ex tracted from the early Belief Revision implementations identified by Doyle and London to inform modern computational implementations of the AGM postulates. Second, it examines whether a unified, knowledge-based formalism can be formulated to govern the structure and dynamics of user profiles. Third, it explores whether the integration of belief revision algorithms can produce user profiles that are more stable, interpretable, and accurate in their evolution than traditional deep learning approaches. The thesis delivers four principal contributions. A systematic literature review of pre-AGM belief revision implementations reveals that early work emphasised representational devices and procedural strategies—such as dependency structures and minimal-change search—that prioritised computational tractability and remain valuable for contemporary systems. The thesis presents the complete and operational implementation of the Knowledge-Driven Profile Dynamics (KDPD) for malism, a modular system that operationalises profiles as integer tuples, employs the Manhattan distance to instantiate the principle of minimal change, and implements all four axiomatically char acterised revision and update operators. A comparative analysis against the NeuroAIReh@b cogni tive rehabilitation framework identifies specific alignments and functional gaps. Finally, a concrete integration pathway maps NeuroAIReh@b workflow steps to KDPD components—encompassing performance-to-formula translation, session-by-session revision, and therapist-driven priority spec ification—thereby establishing a blueprint for deployment in clinical settings. The findings affirm that pre-AGM implementation insights can be successfully integrated with the AGM theoretical framework to yield systems that are both formally grounded and computa tionally viable. The KDPD formalism provides a unified, domain-agnostic framework for profile dynamics that enforces stability through minimal change, enhances interpretability through explicit constraint-driven updates, and offers a principled alternative to opaque data-driven approaches. This work advances the practical application of belief revision theory to user profile management and demonstrates its viability within the domain of cognitive rehabilitation.